2019 Industrial AI Sketch: Automotive Edition

[baoshi.rao]

In 2019, the relationship between AI and the automotive industry could be described as 'a tale of two extremes,' yet amid the frustration, there was also a glimmer of hope. By 2020, both AI and the automotive sector are expected to transition into a warmer spring season.

It's said that the 'Character of the Year' selected this year is 'I'm too south' (a pun on 'I'm too difficult'), coupled with 'no money,' which translates to—

When this phrase is echoed by everyone, it doesn’t seem as bleak anymore.
However, it fits the automotive industry perfectly.
For automotive professionals, 2019 was indeed a year of many heart-stopping moments. Renowned automakers like Daimler, BMW, Audi, General Motors, and Ford announced layoffs, while the emerging electric vehicle sector faced frequent accidents, making even the 'Silicon Valley Iron Man' seem powerless. Progress in autonomous driving and new energy vehicles was slow, with many companies shutting down, leading to a cooling off in investment.

The only bright spot was the rise of intelligent connected vehicles, which garnered significant praise for AI.
From 'making bridal clothes for others' to 'shared prosperity,' what exactly has automotive AI experienced in 2019?
If one word were to describe the relationship between AI and automobiles in 2019, it might be 'a tale of two extremes.'
On one hand, the 'new forces' in car manufacturing, hoping to stage a comeback through intelligence and new manufacturing, saw little success. Companies like NIO and XPeng failed to meet delivery expectations, and their systems were often mocked for bugs. Star autonomous driving projects like Uber, Lyft, Alphabet’s Waymo, Tesla, and Argo also made slow progress, dampening public enthusiasm for the technology.

But if we shift our focus from traditional automakers to the broader transportation sector, we find that AI’s diverse applications are bringing new surprises. For instance, AI is elevating in-car infotainment systems to new heights, transforming how humans interact with vehicles.
In 2018, only autonomous vehicles touted AI features, but by 2019, 'intelligence' became almost a standard in high-end cars.
On one hand, traditional OEMs are adding smart features with deep learning algorithms to enhance Advanced Driver Assistance Systems (ADAS). Meanwhile, startups and tech giants are leveraging their technical strengths to integrate visual algorithms, natural language understanding models, and virtual assistants into automotive systems, creating more personalized driving experiences.

For example, Google and Amazon have embedded smart voice assistants in cars, allowing commands like 'Ok, Google!' and 'Alexa' to control vehicle functions. Sony’s Softkinetic has also developed comprehensive in-car gesture recognition solutions.
For consumers, cars are no longer just a means of transportation but smart devices that integrate various functions to meet diverse needs, much like smartphones.

Driven by such demand, while automakers struggled, suppliers specializing in embedded chips or computational power for vertical automotive applications—such as edge AI hardware semiconductor firms and cloud service providers—achieved remarkable success. For instance, Intel (with Mobileye) and NVIDIA saw significant growth in their computing units.
Google launched Coral AI for edge computing, while NVIDIA released Jetson Xavier NX to provide AI capabilities for cars and other mobile edge devices. Renesas, Xilinx, and Kalray also showcased promising solutions with their specialized products.

While consumer-facing autonomous vehicles made limited progress, commercial autonomous trucks became a favorite among investors in 2019.
Not long ago, Volvo unveiled a new all-electric heavy-duty concept truck and announced plans to disclose financial details of its autonomous driving business starting in 2020, signaling confidence in the sector’s growth.
Meanwhile, TuSimple tested delivery trucks for the U.S. Postal Service, Daimler conducted trials in Virginia, and Starsky Robotics built autonomous fleets using remote teleoperators.

In China, autonomous dump trucks began operating in mines, addressing labor shortages among younger workers. Autonomous taxis and shuttles also appeared in certain urban areas, transitioning from geek toys to familiar public infrastructure.
In 2019, the intelligent transformation of roadside infrastructure for automotive travel also gained momentum, with government-tech collaborations leading the charge.

In July 2019, China’s Ministry of Transport released the 'Digital Transportation Development Plan,' advocating for autonomous driving and vehicle-road coordination technologies, along with dedicated testing sites. Concurrently, 5G-based V2X systems began rolling out in key cities, reshaping how cars interact with urban environments.
Globally, companies are leveraging roadside cameras, public transport sensors, and other tools to redefine mobility. London uses Waze to alleviate downtown congestion, Google’s Sidewalk Labs promotes smart city tech in Toronto, and StreetLight Data tracks traffic patterns using mobile location data.

Overall, applications like remote vehicle diagnostics, traffic efficiency, and contactless payments are enhancing both driving pleasure and travel efficiency. To summarize AI’s role in the 2019 automotive industry: it moved from flashy concepts to deep industrial integration, from singular applications to diverse solutions, and from individual vehicles to large-scale infrastructure.
Predictions often miss the mark, but 2020 is likely to bring surprises as researchers and industry players tackle unresolved challenges, potentially turning lab innovations into real-world breakthroughs.

For example, making cars more 'human-aware.' The unpredictability of autonomous vehicles on real roads, especially regarding human emotions, remains a hurdle for L4+ systems. Integrating social psychology and game theory into algorithms could enhance robustness.
One study addresses AI’s conservative behavior at four-way stops, which, while reducing accidents, frustrates human drivers by slowing traffic. By incorporating 'Social Value Orientation' (SVO), AI can assess driver behavior (selfish vs. altruistic) and adjust its actions dynamically—e.g., taking calculated risks if tailed by an aggressive driver.

Additionally, traditional automakers will likely ramp up AI investments in 2020, shifting focus from vehicle-specific tech to broader applications.
For instance, AI can streamline manufacturing and supply chains. Combining computer vision and predictive algorithms, it can diagnose issues via vibration sensors, preempt machine failures, reduce downtime, and boost profitability.
AI can also optimize R&D priorities, manage budgets, and weed out stagnant projects—a crucial lesson for automakers grappling with losses in 2019.

As vehicles grow smarter, embedded automotive computing systems will soon require upgrades to keep pace.

Currently, most computing is done in the cloud. However, as many in-vehicle controller modules become increasingly intelligent, integrating various decentralized sensing capabilities requires efficient operation from individual AI perception chips. Powerful, low-power, yet more expensive computing units have become the industry's much-needed 'timely rain.' According to Yole data, revenue from automotive infotainment-related computing is expected to grow from $18 million in 2018 to $768 million by 2028.

What new surprises will 2020 bring for automotive-specific coprocessors? This is a question worth pondering and anticipating for the industry.

Over the past year, our expectations for automotive AI have resembled a whale stranded on the beach of technology—frustrating yet filled with longing. As we bid farewell to the 'fiery yet icy' 2019, both AI and the automotive industry are set to be ferried into a warm spring.

In this transitional period of lingering cold and emerging warmth, technology remains the only spark of hope humanity can ignite.